Improvement of the rules selection process in FIS with genetic algorithms

Samir Aliyev, Nigar Ismayilova, Cecilia Zanni-Merk

Abstract: Rule-based solutions for decision-making processes in complex and uncertain environments are beneficial because they are simple, transparent, and Effective. Considering that dependence on expert knowledge and human subjectivity of rule-based systems leads to inconsistencies or inaccuracies, approaches for the automatic rule selection process from training data are critical to minimize problems related to human interference. This study aims to apply genetic algorithms (GA) to automatically select IF-THEN rules in fuzzy inference systems to minimize the problems impacted by human involvement. To demonstrate the proposed approach’s applicability, FIS with an automatized rules selection method based on GA has been applied for the classification of the Titanic disaster dataset. The executed experiments indicate improved classification performance by twice increasing the classification’s F1-score from the initial generation to the last generation. Though the final metrics are less than the current state-of-the-art approach for the given dataset, the results approved the GA’s eligibility for automatic rule selection.